The “Right to Erasure”, additionally known as the “Right to be Forgotten”, is a legal notion that gives people authority over their internet existence. The significance of personal privacy in the digital era is highlighted by this right, which allows people to get their information removed from different online platforms. For machine learning (ML) applications, this means that data must be removed from ML models as well as storage archives. To encounter this issue concept of machine unlearning came into picture. Machine unlearning is growing as a research subject since ML systems must comply with regulations. So generally the requests for removing of a class or part of dataset from a ML model that has already been developed, need to be covered as a part of “Right to be Forgotten”. However, it is not possible to retrain the model from scratch after deleting the data due to computation cost and time duration for retraining the model. Therefore, even for the goal of unlearning, it is frequently the case that no information pertaining to the training procedure or training samples is available. Thus, this poses the question: Can unlearning be accomplished with no training samples and yet achieve efficiency closer to the gold standard? In this work, we present a unique problem: zero-shot machine unlearning through knowledge distillation, which addresses the extreme but realistic case in which there are no usable original data samples. These techniques preserve the model’s effectiveness for the retain data while removing the information from the forget data.